Transport Policy 18 (2011) 102–116

Contents lists available at ScienceDirect

Transport Policy
journal homepage: www.elsevier.com/locate/tranpol

Bus rapid transit impacts on land uses and land values in Seoul, Korea
Robert Cervero a,n, Chang Deok Kang b
a
Department of City and Regional Planning, MC 1850, 228 Wurster Hall, University of California, Berkeley 94720-1850, CA, USA
b
University of Seoul, South Korea

a r t i c l e in f o a b s t r a c t

Available online 23 July 2010 Bus rapid transit (BRT) has gained popularity as a cost-effective alternative to urban rail investments;
Keywords: however, relatively little is known about its impacts on land-use changes and land values. This paper
Bus rapid transit examines the land-market effects of converting regular bus operations to median-lane bus services in
Land use Seoul, Korea, one of the densest, most congested cities in the world. Multilevel models reveal BRT
Capitalization improvements prompted property owners to convert single-family residences to higher density
Hedonic price models apartments and condominiums. Land price premiums of up to 10% were estimated for residences
Multilevel models within 300 m of BRT stops and more than 25% for retail and other non-residential uses over a smaller
Land-use planning impact zone of 150 m. The research findings underscore the importance of introducing zoning and other
Value capture
land regulatory changes prior to the initiation of BRT improvements as well as applying value-capture
tools to help finance investments and redress inequities.
& 2010 Elsevier Ltd. All rights reserved.

1. Introduction exclusive median lanes in 2004 that buses began to offer

More and more cities are turning to bus rapid transit (BRT) as a
way of cost-effectively expanding public transit services to relieve
traffic congestion, reduce carbon emissions, and increase mobility
options for the poor. Because of the inherent flexibility advan-
tages of rubber-tire buses – e.g., unlike rail systems, the same
vehicle that functions as a line-haul carrier can also morph into a
neighborhood feeder – BRT is especially suited for many lower
density and non-CBD settings.
BRT has gained increasing popularity worldwide. Some of the
most advanced and widely heralded BRT services today are found
in Latin America, such as Curitiba and Sa~ o Paulo, Brazil, Bogotá
and Cali, Columbia, Santiago, Chile, and Lima, Peru. The success of
BRT in these cities stems, to a large degree, from the presence of
dedicated lanes, which offer significant speed advantages relative
to more traditional mixed-traffic services. One of the few cities
outside of Latin America that has joined the ranks of world-class
BRT service-providers is Seoul, Korea. As in cities like Curitiba and
Bogotá, Seoul operates dedicated median-lane BRT services which
are supplemented by one of the most extensive networks of
curbside BRT lanes anywhere. Seoul began implementing curbside
bus lanes in 1986; however, because of conflicts with traffic
entering the main traffic stream these lanes failed to provide
significant speed advantages. It was only after the addition of
n
Corresponding author. Tel.: +1 510 642 1695; fax: + 1 520 642 1641.
E-mail addresses: robertc@berkeley.edu (R. Cervero), cdklab@uos.ac.kr
(C. Kang).
significant travel-time savings and win over former motorists.
All else being equal, significant gains in bus speeds should be
followed by significant land-use changes, like densification and
property value increases, especially in congested mega-cities like
Seoul. Land markets can be expected to place a high premium on
parcels close to transit corridors that enjoy significant travel-time
savings since, after all, such settings have scarcity value—i.e.,
there is a finite, limited supply of settings with superior transit
offerings. This paper probes this hypothesis by studying land-use
changes and property-value increases induced by Seoul’s intro-
duction of exclusive, median-lane BRT services. First, the
empirical literature on bus transit and land-use impacts is
reviewed. This is followed by background discussions on Seoul’s
transportation conditions and BRT system. Next, we describe our
research methodology and supporting data sources. We then
present multilevel models that gauge the influences of upgrading
BRT services on land-use changes and land values. The paper
concludes by reflecting on the policy implications of the key
research findings.
2. Literature review
A large body of literature confirms that urban real-estate
responds positively to transportation improvements, mainly in
the form of higher property values and, zoning permitting, land-
use intensification (Cervero, 1997; Ryan, 1999). Transportation
infrastructure increases the supply of developable land and

0967-070X/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tranpol.2010.06.005
 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116
through the competitive bidding process increases the price of
land for parcels that enjoy significant gains in accessibility
(Dowall and Monkkonen, 2007; Ewing, 2009). The benefits of
new transportation investments get capitalized in real estate
prices in the short-term while over the longer term land-use
adjustments occur. Thus while land-price impacts can be
instantaneous, land-use changes tend to be slower, partly due to
institutional lags (e.g., in securing building permits and zoning
amendments) (Perez et al., 2003).
Most transportation capitalization studies to date have focused
on highway corridors in the developed world. Given the
predominance of automobile travel in countries like the United
States, not surprisingly larger value gains have been recorded as a
consequence of highways improvements vis-a -vis expanded or
enhanced transit services (Cervero, 1997; Ryan, 1999; Bhatta and
Drennan, 2003). Studies generally find, however, that the impacts
of highways on land-use changes are largely redistributive,
shifting growth that might otherwise occur in some settings to
newly served highway settings (Cambridge Systematics et al.,
1998; Boarnet, 1998; Boarnet and Chalermpong, 2001). Boarnet
and Haughwot (2000) note that redistribution need not be a zero-
sum situation because economic benefits can accrue from spatial
relocation, such as increasing agglomeration economies. In
addition, other preconditions—such as deregulated zoning, higher
permissible densities, and infrastructure provisions—are often
required if road investments are to have significant impacts on
land use (Giuliano, 2004). Regional location also matters. Research
shows that land-price adjustments tend to be smaller as the
distance of a highway to the CBD increases (Voith, 1993).
Most studies of public transit’s impacts on cities and land
values have focused on heavy rail systems since such capital-
intensive investments have historically conferred the most
significant accessibility benefits of any transit improvements.
Much of this research has focused on recent-generation rail
systems built in US cities over the past half-century (Huang, 1996;
Giuliano, 2004). Empirical research on rail investments and land-
price impacts has produced mixed results. Studies of San
Francisco’s BART found considerable variation in land-price
impacts, with downtown San Francisco commercial properties
reaping huge gains and many suburban residential settings
experiencing no discernible impacts (Cervero and Landis, 1997).
Research on Miami’s Metrorail recorded no significant land-price
effects owing to low ridership and flat real-estate markets in
many areas that were served (Gatzlaff and Smith, 1993). A
study of Chicago’s Midway Line showed that the opening of
new rail services increased housing prices, with rates of
land-value appreciation varying over time (McMillen and
McDonald, 2004).
Outside the US, a range of impacts have also been found,
reflecting the highly contextualized nature of rail-induced land-use
impacts. In Seoul, Korea, residential property values were found to
have risen before a new subway line opened, reflecting the
speculative nature of rail-induced land appreciation in a dense,
crowded city. The proximity effects of rail on land values, however,
were less than that of other predictors such as housing character-
istics and distance to recreational facilities (Bae et al., 2003). Because
they have a longer history of urbanization that predates the
automobile age, new rail investments in European cities are widely
viewed as strengthening and reinforcing compact, pedestrian-scale
development patterns (Banister and Lichfield, 1995; Hall, 1995;
Banister, 2005). The Royal Institution of Chartered Surveyors (2002)
surveyed 150 studies drawn mainly from the UK and continental
Europe. The study concluded that public transit services generally
exerted positive influences on residential as well as commercial
properties; however, because of large statistical variation, a host of
intervening factors likely affect outcomes as well.
103
Conventional wisdom holds that traditional bus transit
services have imperceptible influences on urban form and land-
use patterns because, in contrast to many rail systems, they fail to
confer appreciable accessibility benefits. This is especially the case
in the developed world where high levels of private automobile
ownership means conventional buses are considerably slower
than cars for the vast majority of trips. The exception to this rule,
however, could be BRT wherein buses are provided with an
exclusive, dedicated lane, signifying a significant improvement in
service quality in the minds of real-estate developers and
property owners (Polzin and Baltes, 2002). Levinson et al.
(2002) contend that BRT investments in Ottawa, Pittsburgh,
Brisbane, and Curitiba generated land-use benefits that were as
large as those that would have been created by railway
investments. Vuchic (2002) expresses doubt, arguing that light-
rail transit (LRT) has a significantly higher potential to impact
urban form than BRT.
Empirical evidence that might inform this debate is quite
limited. Several past studies have investigated the affects of BRT
on land values. A study of dedicated-lane BRT services in Los
Angeles found small negative impacts on residential property
values and small gains for commercial parcels (Cervero, 2004).
Land-value impacts of light-rail services in Los Angeles were
found to be similar to those of BRT—i.e., slight declines in
residential values and fairly small gains in commercial properties
(smaller than that found for BRT). In contrast, a study of the more
substantial BRT system in Bogotá, Colombia found appreciable
land-value benefits. There, multi-family housing units close to
Bogotá’s TransMilenio BRT rented for more per square meter than
units located farther away (Rodriquez and Targa, 2004). The
addition of new TransMilenio lines was found to further increase
the land-value premium of already-served properties (Rodriquez
and Mojica, 2008). Better pedestrian environments around
Transmilenio stops also contributed to higher land values
(Estupinan and Rodriguez, 2008). In the case of Brisbane,
Australia, the southeast busway, which as in Bogotá operates on
an exclusive lane, was similarly found to confer land-value
benefits to nearby residential properties (Levinson et al., 2002).
As with rail, various mediators can influence the land-value
impacts of bus-based investments. Heterogeneous neighborhood
attributes can alter how changes in transit travel times affect
residential property prices (Du and Mulley, 2006). Clustering
transit stops in commercial and mixed-use districts can also yield
measurable land-value benefits (Zhou and Kockelman, 2008).
Our research aims to extend insights gained from previous
analyses by examining changes in both land use compositions and
land values following BRT improvements over several time points.
We investigate impacts on both residential and non-residential
properties along affected BRT corridors in Seoul. As background to
the study, the next section describes Seoul City and its current
BRT services.
3. Background information on BRT in Seoul
Seoul is the capital of Korea and the nation’s economic,
political, and cultural hub. The city itself, with more than 10
million inhabitants, is part of the Seoul Metropolitan Area (which
includes Kyunggi Province and Incheon city), the world’s second
largest conurbation at 23 million (Fig. 1). With 16,000 residents
per square kilometer, Seoul and Incheon comprise the sixth
densest urbanized area in the world (Fig. 2).
Economic growth and rapid urbanization have brought about
steady increases in car ownership and congestion levels in Seoul
(Fig. 3). Between 1995 and 2005, average motor-vehicle speeds in
Seoul hovered around 20–25 km/h, with the worst congestion
104 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116
during evening peak hours (Fig. 4). Partly because of extreme
traffic congestion as well as for income reasons, the majority of
Seoul residents travel by public transport. From 2003 to 2006,
more than 60% of motorized trips were by bus or subway (Fig. 5).
Because of transit’s shrinking modal shares and worsening
traffic congestion various bus-transit reforms were introduced in
the mid-1990s, including the provision of dedicated curbside bus
lanes. These improvements failed to stem bus-transit’s secular
declines in ridership as its modal shares fell from 30% in 1996 to
26% in 2002 (while subway’s share rose from 29% to 35% during
the same period). Part of the reason for bus-transit’s decline was
excessive competition among private operators which resulted in
Fig. 1. Location of Seoul, Korea.
Source: Seoul Metropolitan Government.
Fig. 2. Rank order of population densities among global cities (2006).
Source: City Mayors (www. citymayors.com).
redundant and unstable services, the skipping of stops, and overly
aggressive driving. These factors, along with rising operating
deficits, prompted the Seoul Metropolitan Government to
introduce a semi-public transit organization in the early 2000s
that set and enforced rules and standards on bus routes,
schedules, and private operating practices. Many bus routes were
reorganized into a timed-transfer and pulse-scheduling arrange-
ment. Moreover, all bus services were classified into four types of
colored services: red (long-distance and intercity services), blue
(trunk services), green (feeder services), and yellow (circular
services). The red long-distance intercity lines linked satellite
cities with each other and downtown Seoul while blue trunk lines
connected between the sub-core and central-city Seoul. Green
feeder buses mainly funneled passengers to subway stations and
express bus stops. Yellow circular lines orbited the urban core.
Equally important was the full-scale upgrade of BRT services.
During the early 2000s, Seoul’s curbside bus lanes were expanded
from 219 to 294 km, and in mid-2004 dedicated median-lane
services were introduced (Fig. 6). By 2008, Seoul had installed
74 km of median-lane BRT services spanning 8 corridors (Fig. 7).
The combination of dedicated lane-services, bus-priority traffic
signals, real-time passenger information systems, and attractively
designed bus stops materially improved service quality.
Six months after the introduction of median-lane bus services,
average bus operating speeds doubled from 11 to 22 km/h (Seoul
Development Institute, 2005a). Table 1, which compares bus versus
car speeds along three road segments of the BRT network, shows
bus users enjoyed substantial travel-time savings relative to
motorists. Other benefits included a reduction in bus-related
accidents and improved schedule adherence. As a consequence,
previous declines in bus transit’s ridership were reversed, with bus
patronage jumping 10% between the end of 2003 (prior to median-
lane services) and the end of 2004 (after median-lane services).
These ridership gains have been sustained: in 2009, bus-transit
patronage outnumbered that of the subway system by more than
100,000 daily passengers; six years earlier, subways carried nearly
a million more passengers per day than buses (Seoul Metropolitan
Government, 2009). Not surprisingly, passenger satisfaction
increased following the introduction of median bus-lanes in
2004, as shown in Table 2. And there was a clear association
between where people lived and level of satisfaction. A survey of
3000 passengers in November 2004 revealed that 28% were
satisfied with overall bus service improvements. However, among
 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116
Fig. 3. Registered motor vehicles in Seoul (1995–2005).
Source: Seoul Metropolitan Government.
Fig. 4. Average speeds of motor vehicles, 1995–2005.
Source: Seoul Metropolitan Government.
those living in districts with exclusive median bus lanes, more than
half said they were very satisfied with changes (Seoul Development
Institute, 2005b). Fig. 8 reveals the strong spatial association
between where satisfied residents lived and the location of
median-lane bus services.
In parallel to improvements in BRT services were a number of
other reforms introduced under the leader of Myung-Bak Lee,
former mayor of Seoul and now president of South Korea who
supported a more transit-oriented built form. One was an
ambitious campaign of land reclamation, taking valuable
central-city real estate given over to the private car and
transforming parcels into attractive public spaces. Most notice-
able was the removal of a 6-km elevated freeway in the heart of
Seoul, Cheong Gye Cheon (CGC), replaced by a restored urban
stream and pedestrian-friendly greenway. Mayor Lee also con-
verted a 1.3 ha surface-street intersection in front of Seoul’s City
Hall with an oval-shaped grass park. Furthermore, in reaction to
growing public discontent over excessively long commutes
between far-flung new towns and central Seoul, local government
embarked on a New Town-In Town program. Seoul’s city
government sought to jump-start central-city redevelopment by
providing various public amenities like green space and expand-
ing infrastructure and public services. Many of these ‘‘Promotion
Areas’’ were sited along the median-lane BRT corridors (Fig. 9).
105
4. Research methodology and data sources
To study the effects of Seoul’s 2004 BRT reforms on land-use
activities and property values, we gathered parcel-level data for
affected properties over multiple time points. Since land use is
measured on a nominal scale, logit models were used to gauge the
influences of BRT on discrete land-use changes. For studying
impacts on the ratio-scale variable, assessed land value, we
applied multiple regression techniques, specified according to
hedonic price theory (Rosen, 1974). Hedonic price models
apportion land-price effects based on the attributes of buildings
and land as well as characteristics of surrounding neighborhoods.
Since such attributes are measured at different geographic scales
(e.g., parcels versus neighborhoods), as discussed later, multilevel
techniques were used to estimate best-fitting models.
Numerous data sources were drawn upon to probe the land-
use and land-value impacts of Seoul’s improved BRT services.
Table 3 lists and describes the key variables that were collected as
well as data sources. Particularly important were data obtained
from annual land surveys conducted by the Seoul Assessor Office
from 2001 to 2007. For each parcel in the city, this survey
provided information on street address, land use, assessed
land value, and other features. Land-value data were adjusted
using a consumer price index (CPI) to control for inflation effects
over time.
The sample frame for our research comprised land parcels
whose nearest bus stop became a median-lane stop once the BRT
improvements were introduced in 2004. Thus if a parcel was
closer to a median-lane bus stop in 2004 than a regular bus stop, it
was included in our sample; if it was closer to a regular bus stop,
it was not. This yielded more than 187,000 parcel observations
(the majority of which were residential properties) for model
estimation. All parcels were within 2150 m of a BRT stop and the
vast majority were within half kilometer.
With land-parcel data in hand, point-based maps were then
created to measure network and straight-line distances from each
surveyed parcel to the nearest BRT stops (shown in Fig. 7) as well
as to major roads, subway stations, the Han River (Seoul’s major
north-south dividing line), and as a hub of Seoul’s ambitious land
reclamation and redevelopment campaign, the Cheong Gye Cheon
(CGC) corridor. As Fig. 10 shows, CGC was a central link in Seoul’s
thoroughfare network and its demolition placed demands on
106 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116
Fig. 5. Modal shares in Seoul, 2003–2006.
Source: Seoul Metropolitan Government.
Fig. 6. Bus median lanes in Seoul.
Source: Seoul Metropolitan Government.
alternative services, including BRT, to absorb displaced traffic. The
very existence of median-lane BRT was due, in part, to the lost
capacity from the freeway’s demolition, thus the spatial
relationship of studied parcels to not only BRT stops but also
the CGC corridor was of interest.
Fig. 7. Map of BRT corridors in Seoul.
Source: Adapted from Seoul Metropolitan Government.
Since our models relied on information from neighborhoods
that surrounded surveyed parcels, various socio-economic vari-
ables were also compiled, as shown in Table 3. Statistically, these
variables served as controls, allowing us to partial out the unique
effects of proximity to median-lane BRT stops on land-use and
land-value changes. A variable like ‘‘park ratio’’ (a proxy for the
amount of open space and greenery in an area), for instance, could
be expected to increase residential property values in a crowded,
congested city like Seoul. Such variables should be included in a
hedonic price model to statistically remove potential confounding
effects.
5. Land use change models
This section examines how Seoul’s introduction of median-lane
BRT improvements in mid-2004 affected land uses. Multilevel binary
logit models were used to predict three types of conversions from
 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116 107

Table 1
Comparison of operating speeds (km/h) of cars and buses along three road segments with exclusive median bus lanes.
Source: Seoul Development Institute (2005a).

Description Before (June 2004) After (August 2004) Percentage change

Road A Bus (exclusive lane) 11 20.3 85.0

Car (other lane) 18.5 19.9 7.6
Road B Bus (exclusive lane) 13.1 22.5 72.0
Car (other lane) 20.3 21 3.4
Road C Bus (exclusive lane) 13 17.2 32.0
Car (other lane) 18 19.1 6.1

Table 2
Number of formal public complaints about bus services, before and after median-lane BRT services and other service reforms.
Source: Seoul Development Institute (2005a).

Type of complaints April 2004 (before) December 2004 (after) May 2005 (after)

Transport card and fare 59,871 4820 640

Service routes 1216 44 15
Service schedules 1638 141 29
Bus stops, route maps 561 24 4
Service for bus driver 392 40 30
Publicity of route and fare 331 19 1
Other (suggestion, transfer) 981 48 34
Total 64,990 5136 753

Fig. 8. Percentage of satisfaction and location of bus stops.

Source: Seoul Development Institute (2005b).

single-family residences: to multi-family residential rental units, to
condominium owner-occupied units, and to mixed-parcels which
typically involved a combination of commercial activities (e.g., retail,
services, offices) and sometimes residential as well. All of these
changes correspond to what might be considered an intensification
of activities on parcels, from single-family residences to often higher
density activities (i.e., more units in the form of multi-family
housing and condominiums; adding of retail activities). To the
property owner, intensification normally translates into higher
valued properties and in some cases increases in rental income.
We note that there were insufficient observations to model other
Fig. 9. Location of new towns-in town and promotion areas along BRT lines.
Source: Adapted from Seoul Metropolitan Government.
possible land-use conversions, such as from retail-to-offices or
undeveloped-to-retail.
The land use statuses of more than 52,000 single-family
residential parcels that were part of our sample frame were tracked
for the 2001–2007 period. More than 96% of parcels remained in
single-family use over this six-year period. Among the remaining
parcels, the dominant conversion was to multi-family housing
followed by mixed land uses and condominiums. Fig. 11 shows the
locations of converted parcels, all aligned fairly close to BRT stops.
108 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116

Table 3
Variables and data sources for modeling land-use and land-value impacts.

Variables Description Data source

Dependent Variables
CPI-adjusted land value (Korean Won/m2) Land value adjusted with CPI (2005¼ 100) Annual Land Survey
Land Use Change Types Selected land use change ¼1, No change ¼ 0 Annual Land Survey
Independent Variables
Location factors
Distance to nearest CGC freeway ramps Straight-line distance to CGC elevated freeway ramp(m) Calculated using GIS
Network distance to nearest CGC freeway ramps Distance along network to CGC elevated freeway ramps(m) Calculated using GIS
Straight-line distance to pedestrian entrances on CGC urban
Distance to nearest CGC pedestrian entrances Calculated using GIS
greenway(m)
Network distance to nearest CGC pedestrian Distance along network to pedestrian entrances on CGC urban Calculated using GIS
entrances greenway(m)
Distance to CBD: City Hall Straight-line distance to Seoul’s City Hall(m) Calculated using GIS
Distance to nearest subway station Straight-line distance to nearest subway station(m) Calculated using GIS
Distance to nearest urban arterial Straight-line distance to urban arterial(m) Calculated using GIS
Distance to bus stop Straight-line distance to bus stops(m) Calculated using GIS
Network distance to bus stop Distance along network to bus stops(m) Calculated using GIS
Distance to Han River Straight-line distance to Han River(m) Calculated using GIS
Job accessibility within 30 min by Car Number of jobs within 30 min by car Calculated using GIS

Land attributes, use, and regulation

CPI-adjusted land value (Korean Won/Square
Meter) Land value adjusted with CPI (2005¼ 100) Annual Land Survey
Land use Land use types (residential and non-residential) Annual Land Survey
Building coverage ratio Ratio of floor area to total land area Seoul Zoning Map
Floor area ratio Ratio of total building area to floor area Seoul Zoning Map

Neighborhood economic and demographic attributes

Population density Number of residents per total district area Seoul Statistics
Employment density Number of jobs per gross Ward (neighborhood) area Seoul Statistics
Proportion in age groups 20–40, 40–60, and more than 60 per
Age structure Seoul Statistics
resident over 20 years of age
Proportion of residents with college degree per resident over 20 Population and Housing
Proportion of residents with college degree
years of age Census

Other neighborhood attributes

Park ratio Park area per gross Ward (neighborhood) area Seoul Statistics
Land for building, school, and road per gross Ward (neighborhood)
Developed land ratio Seoul Statistics
area
Road area ratio Total road area per gross Ward (neighborhood) area Seoul Statistics
Retail area ratio Total retail building area per gross Ward (neighborhood) area Seoul Statistics
Proportion of residential permits to total permits Total area of residential permit per gross permit area Seoul Statistics
Proportion of commercial permits to total permits Total area of commercial permit per gross permit area Seoul Statistics
CPI-adjusted local tax per households (Korean Won) CPI-adjusted local tax per households (2005¼ 100) Seoul Statistics

5.1. Model structure
Multilevel logit models were used to estimate factors influencing
the three land-use conversions. Multilevel modeling accounts for the
fact that parcels from the same neighborhood share common
attributes, like local road-network designs and demographic
characteristics. Failure to account for shared upper-level (i.e.,
neighborhood) attributes of lower-level (i.e., parcel) observations
can bias parameter estimates (Raudenbush and Bryk, 2002). Our
multilevel models incorporated both fixed and random effects. Fixed
effects represent variable coefficients that are constant across upper-
level (i.e., neighborhoods) units while random effects indicate error-
terms that vary across upper level units. Estimated multilevel
models of land-use conversion took the following form:
 
yij
log it ¼ g00 þ bK LijK þ bK SijK þ bK NijK þ zj þ eij ð1Þ
1yij
where yij ¼1 if single-family parcel i (Level 1) in neighborhood j
(Level 2) changed use; 0 if no change of land use; g00 is the model
constant; bK are the coefficients of variables (k¼1, 2, 3,y,m,
m¼number of variables), LijK is a vector of location attributes (e.g.,
distance to bus stops) of parcel i (Level 1) in neighborhood j (Level
2); SijK is a vector of neighborhood socio-demographic characteristics
(e.g., population density, education level) in neighborhood j (Level 2)
for parcel i (Level 1) in that neighborhood; NijK is a vector of
neighborhood land-use (e.g., share of parcels in retail use) and
public expenditure (e.g., local tax per household) in neighborhood j
(Level 2) that is assigned to each parcel i (Level 1) in the
neighborhood; and zj, eij are the residual error terms of Level 2
and Level 1, respectively.
An important output of multilevel models is the intraclass
correlation (ICC), which measures the relative variation in the
estimated dependent variable between versus within neighbor-
hoods. High ICC values, typically above 0.05 and with statistically
significant probability levels, indicate individual parcels tend to
share neighborhood attributes, signifying the need for multilevel
estimation (Rabe-Hesketh and Skrondal, 2008).
5.2. Model results
Table 4 presents the multilevel model results for the most
dominant conversion – single-family to multi-family residential
– and Table 5 shows the output for the two other land-use changes
studied: single-family to condominiums and to mixed uses. Slightly
better model fits were obtained when expressing ratio-scale
explanatory variables in natural logarithmic form, thus these
model results are presented. Models were specified according to
the multilevel structure described earlier in Eq. (1).
Of most interest to our research is the affects of ‘‘distance to bus
stops’’ on land-use conversions. For all single-family parcels in the
 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116
Fig. 10. Urban arterials and freeways with reference to CGC corridor.
Source: Adapted from Seoul Metropolitan Government.
sample frame, Tables 4 and 5 reveal that parcels within half
kilometer of a stop (generally associated with a walk of under 5 min)
were generally more likely to convert to more intensive uses relative
to parcels beyond half kilometer. Impacts across 100 m distance
bands did not follow a simple linear pattern, as plotted in Fig. 12,
suggesting other intervening and unaccounted-for factors, such as
urban design features, might have influenced distance effects. We
note that higher-end conversions – to condominiums and mixed-use
buildings – were actually less likely to occur in the immediate
vicinity of a bus stop (i.e., o100 m). This could be due to the
nuisance effect of being located near busy BRT and roadway
corridors (e.g., people walking to and congregating around bus
stops; noise impacts). Multi-family conversions, however, seemed
immune to such possible nuisance effects. Beyond a buffer distance
of 100 m to a stop, single-family conversions were more likely to
occur. At around 400 m, the influences of distance to a bus stop on
land-use conversions largely evaporated.
Among other location variables, distance to arterial roads had
the strongest influence on land-use conversions; the likelihood of
switching to multi-family and mixed uses fell with distance to
arterial roads. Other distance variables (e.g., to city hall and
subways) were statistically associated with condominiums and
mixed uses conversions, albeit in no clear or discernible pattern.
Among the remaining variables, higher assessed land values of
a neighborhood significantly increased the odds of converting
single-family residences to the higher end uses: condominiums
and mixed uses. Property owners seemed particularly inclined to
convert residences to condominiums, the most popular high-rise
housing in Korea, in settings with relatively high average land
values as well as college-educated residents. Less appealing in
109
higher valued core areas of the city were conversions to multi-
family housing. Table 5 also shows that higher permissible floor
area ratios of a neighborhood contributed to mixed-use conver-
sions. Most other control variables in Tables 4 and 5 were not
statistically significant but were retained because their inclusion
is consistent with hedonic price theory (Rosen, 1974; Bartik,
1988) and also to apply consistent sets of explanatory variables
across all models. Additionally, because of the large size of our
sample, the influence of statistically insignificant control variables
should have no discernible effects on the efficiency of other
parameter estimates (Stigler, 2008).
6. Land value models
A similar multilevel modeling approach was adopted for
studying the land-value capitalization effects of Seoul’s BRT
improvements. The primary change to Eq. (1), shown earlier,
was the use of assessed land values, a ratio-scale variable, as the
left-hand-side dependent variable. The multilevel regression
model conceptually is similar to multilevel logit model, presented
in Eq. (1), except binary logit models are estimated for nominal
dependent variable (Rabe-Hesketh and Skrondal, 2008).
Using land valuation data from Seoul’s Assessors’ Office,
multilevel multiple regression models were estimated for resi-
dential and non-residential properties over two time periods:
2001–2004 (pre-BRT) and 2005–2007 (post-BRT). Since land
values are assessed in Seoul in the early part of the calendar
year, the 2004 valuations were made prior to the mid-2004
initiation of median-lane BRT services. Before-and-after data
should allow the influences of BRT improvements on land values
to be gauged. Our analysis thus assumes there was little
anticipatory land-value appreciation prior to the BRT improve-
ment. Because BRT in South Korea was untested and untried in
2004, many local observers were skeptical it would deliver
benefits. Indeed, the project faced considerable opposition
because of concerns over possible increases in traffic congestion,
crowding on buses, and lost retail sales to business merchants. For
these reasons, we assumed capitalization benefits might only
have accrued once the service was in actual operations.
Multilevel regression models on land-value impacts were
estimated in log–log form because they produced better statistical
fits than linear models and also moderated the effects of
heteroschedastic error terms. A side benefit of log–log formula-
tions is that estimated coefficients represent elasticities, revealing
the relative sensitivity of land values to changes in the right-hand
side predictor variables. The models shown in this section also
produced the best statistical fits and were consistent with hedonic
price theory.
6.1. Non-residential hedonic price models
A total of 37,515 and 23,969 non-residential land-parcel
observations were available, respectively, for the two study
periods: 2001–2004 and 2005–2007. For both periods, non-
residential parcels in our sample were comprised of the following
land uses: commercial-retail (55.2%), office (3.3%), undeveloped
land zoned for commercial-retail (3.1%), mixed-use (37.6%), and
undeveloped land zoned for mixed-use (0.7%). The parcels ranged
in value per m2 from 637,000 Korean Won (US$500) to 45 million
Korean Won (over US$35,000) over the 2001–2007 period.
The multilevel hedonic-price models estimated for non-
residential parcels are shown in Table 6. The intraclass
correlations, indicating the share of variation explained by the
grouping structure, were quite high, justifying the use of
multilevel model estimation. Notably, 72.2% and 88.9% of the
110 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116

Fig. 11. Location of converted single-family residential parcels.

Source: Adapted from Seoul Metropolitan Government.
 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116 111

Table 4
Multilevel logit model for predicting single family housing to multi-family conversions.

SF to Multi-family housing
Variables Coefficient t p

Fixed effects
Distance to bus stops
dummy (1, if distance r 100 m, otherwise 0) 1.253 2.320 0.020
dummy (1, if 100 o distance r 200 m, otherwise 0) 1.657 3.150 0.002
dummy (1, if 200 o distance r 300 m, otherwise 0) 1.699 3.290 0.001
dummy (1, if 300 o distance r 400 m, otherwise 0) 1.999 3.920 0.000
dummy (1, if 400 o distance r 500 m, otherwise 0)  0.120  0.190 0.851

Location factors
ln(Network distance to CGC pedestrian entrances) 0.078 0.130 0.898
ln(Distance to CBD: City Hall) 0.900 1.300 0.194
ln(Distance to nearest subway station) 0.032 0.350 0.726
ln(Distance to nearest urban arterial)  0.130  3.450 0.001

Neighborhood economic and demographic attributes

ln(CPI-adjusted land value)  1.462  9.950 0.000
ln(Population density) 0.607 2.410 0.016
ln(Employment density)  0.661  0.380 0.703
ln(Proportion of residents with college degree) 1.233 2.500 0.012
ln(Proportion of residents 40–60 years of age) 0.766 0.490 0.622
ln(Proportion of residents more than 60 years of age) 0.352 0.220 0.823

Other neighborhood attributes

ln(Park density ratio)  0.349  1.220 0.223
ln(Developed land ratio) 1.778 1.240 0.214
ln(Road area ratio)  0.897  0.290 0.774
ln(Retail area ratio)  0.233  1.210 0.226
ln(Proportion of residential permits to total permits) 0.241 0.780 0.438
ln(Proportion of commercial permits to total permits) 1.010 1.260 0.207
ln(CPI-adjusted local tax per households) 0.859 0.790 0.428
ln(Job accessibility within 30 min by Car)  0.395  0.630 0.526

Constant 1.846 0.110 0.910

Random effects
Standard deviation of the random intercept 0.718
ICC 0.136
Summary statistics
Number of parcel observations (Level 1) 25,410
Number of neighborhood groups (Level 2) 72

variation in land values is explained by between-group variation
among 71 and 70 neighborhoods over the two time periods,
respectively. Most predictor variables in the models shown in
Table 6 were statistically significant at the 5% probability level.
The coefficients on dummy variables for distance to the nearest
BRT stops speak to the core research question: whether proximity
affects land prices differently before and after the BRT improve-
ments. Fig. 13 plots these coefficients, revealing the marginal effects
of proximity on land prices, expressed in percentage terms and over
30 m distance bands, relative to parcels more than 300 m away.
While there were general proximity benefits in both periods, Fig. 13
reveals the benefits were more prominently capitalized into land
values in the post-period (2005–2007). Impacts were particularly
notable within 150 m of the nearest bus stop. The lack of statistical
significance of dummy variables for locations beyond 150 m of a BRT
stop in the post-period suggests BRT services produced highly
localized land-market premiums.
Other control variables in Table 6 generally conform to
expectations. Land prices fell with distance to the nearest CGC
freeway ramps (when they existed in 2001–2004) as well as the
nearest CGC greenway pedestrian entrances (in the post-freeway
period of 2005–2007). They also fell with distance to another
important infrastructure component, Seoul’s world-class subway
system. Table 6 also shows offices enjoyed higher land-value
premiums than other non-residential uses, ceteris paribus, and site
density (as reflected by building coverage and floor-area-ratio)
also worked in favor of higher land values (though only in the pre-
BRT period). While the signs on some control variables, such as
‘‘park density ratio’’ (reflecting the benefit of having parks in the
neighborhoods), make sense, the signs on others are less easy to
explain and likely reflect local idiosyncrasies of Seoul’s commer-
cial real-estate market. As previously noted, the very large size of
our sample minimizes any unintended effects of statistically
insignificant control variables on other parameter estimates
(Stigler, 2008).
6.2. Residential hedonic price models
In all, data for 85,124 and 41,302 residential parcels were
available for the two periods. For both periods, residential parcels
were used as follows: single-family housing (81.9%), multi-family
housing (11.5%), undeveloped land zoned for residential (3.7%),
condominiums (1.7%), and row housing (1.3%). Residential parcels
ranged in value per m2 from 148,000 Korean Won (US$32) to
more than 8,400,000 Korean Won (US$6600) over the 2001–2007
period.
The multilevel models estimated for residential parcels are
shown in Table 7. The high intraclass correlations justified the use
of multilevel model estimation: 99.2% and 99.5% of the variation
in land values is explained by the between-group variation across
the 65 neighborhoods over the two time periods, respectively.
Nearly all predictor variables in both models are statistically
significant at the 0.01 probability level. As with the non-
residential model, distance to the nearest BRT stop had a
significant and discernable effect on residential land prices,
112 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116

Table 5
Multilevel logit model for predicting single family housing to condominium and mixed-use conversions.

SF to condominium SF to mixed-use
Variables Coefficient t p Coefficient t p

Fixed effects
Network distance to bus stops
dummy (1, if network distance r 100 m, otherwise 0)  28.826 0.000 1.000  1.185  2.890 0.004
dummy (1, if 100 o network distance r 200 m, otherwise 0) 0.173 0.310 0.754 0.024 0.110 0.913
dummy (1, if 200 o network distance r 300 m, otherwise 0) 1.023 2.370 0.018 0.431 2.170 0.030
dummy (1, if 300 o network distance r 400 m, otherwise 0) 0.565 1.450 0.147 0.541 2.740 0.006
dummy (1, if 400 o network distance r 500 m, otherwise 0) 0.342 0.900 0.367  0.087  0.390 0.698

Location factors
ln(Network distance to CGC pedestrian entrances) 7.127 2.290 0.022 0.959 1.270 0.204
ln(Distance to CBD: City Hall)  22.832  4.940 0.000  1.310  1.770 0.077
ln(Distance to nearest subway station) 0.805 2.340 0.019 0.462 3.720 0.000
ln(Distance to nearest urban arterial) 1.112 6.060 0.000  0.262  4.830 0.000
ln(Distance to nearest bus stop) 1.271 4.070 0.000

Neighborhood economic and demographic attributes

ln(CPI-adjusted land value) 2.310 6.540 0.000 0.609 4.040 0.000
ln(Building coverage ratio)  0.297  0.430 0.665
ln(Floor area ratio) 0.411 2.600 0.009
ln(Population density)  7.614  3.230 0.001 0.053 0.170 0.867
ln(Employment density)  46.629  0.030 0.976 3.495 1.280 0.199
ln(Proportion of residents with college degree) 12.475 2.140 0.032 0.602 0.930 0.353
ln(Proportion of residents 40–60 years of age)  22.523  1.500 0.134  0.826  0.390 0.697
ln(Proportion of residents more than 60 years of age)  46.801  2.260 0.024  5.827  2.840 0.005

Other neighborhood attributes

ln(Park density ratio)  0.351 0.000 0.999 0.080 0.230 0.816
ln(Developed land ratio)  106.385  0.030 0.976  0.172  0.100 0.922
ln(Road area ratio) 95.790 0.030 0.979  3.801  0.850 0.393
ln(Retail area ratio) 2.598 0.010 0.990 0.505 1.440 0.149
ln(Proportion of residential permits to total permits) 13.544 0.040 0.968 0.723 1.460 0.144
ln(Proportion of commercial permits to total permits)  20.038  0.020 0.984  0.721  0.850 0.396
ln(CPI-adjusted local tax per households) 22.288 0.010 0.991  2.054  1.340 0.179
ln(Job accessibility within 30 min by car)

Constant 277.969 0.020 0.983  28.466  1.610 0.108

Random effects
Standard deviation of the random intercept 4.886 1.002
ICC 0.879 0.234
Summary statistics
Number of parcel observations (Level 1) 2,387 24,810
Number of neighborhood groups (Level 2) 65 72

10.00
5 00
5.00
0.00
-5.00
Coefficiients

-10.00
SF to Multi Family
-15.00
Housing
-20.00 SF to Condominium
-25.00 t Mi
SF to dU
Mixed-Use

-30.00
-35.00
0~100 100~200 200~300 300~400 400~500
Network Distance to Bus Stops (m)

Fig. 12. Coefficients of each land use change by distance intervals.

underscored by Fig. 14. The figure shows residential land prices bigger once median-lane BRT services were introduced. Prior to
were generally higher for parcels within 300 m of a bus stop than these services, residential values were slightly lower within 30 m
those beyond 300 m; however, the premium effect was noticeably of the nearest bus stop, likely reflecting a nuisance effect, and then
 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116 113

Table 6
Multilevel hedonic model for predicting non-residential land value per square meter. Note: 1 Korean Won¼0.0011 US dollar in 2007.

2001–2004 2005–2007

Variables Coefficient t p Coefficient t p

Fixed effects
Distance to bus stops
dummy (1, if distance r 30 m, otherwise 0) 0.183 14.440 0.000 0.260 16.390 0.000
dummy (1, if 30 o distance r 60 m, otherwise 0) 0.137 11.790 0.000 0.189 13.160 0.000
dummy (1, if 60 o distance r 90 m, otherwise 0) 0.047 4.110 0.000 0.096 6.790 0.000
dummy (1, if 90 o distance r 120 m, otherwise 0) 0.045 3.950 0.000 0.061 4.290 0.000
dummy (1, if 120 o distance r 150 m, otherwise 0) 0.031 2.730 0.006 0.033 2.290 0.022
dummy (1, if 150 o distance r 180 m, otherwise 0) 0.022 1.970 0.049 0.001 0.040 0.969
dummy (1, if 180 o distance r 210 m, otherwise 0) 0.033 2.810 0.005  0.003  0.210 0.837
dummy (1, if 210 o distance r 240 m, otherwise 0) 0.061 4.940 0.000 0.008 0.510 0.609
dummy (1, if 240 o distance r 270 m, otherwise 0) 0.045 3.430 0.001 0.008 0.470 0.640
dummy (1, if 270 o distance r 300 m, otherwise 0) 0.040 2.610 0.009 0.002 0.090 0.928

Location factors
ln(Network distance to nearest CGC freeway ramps)  0.804  23.250 0.000
ln(Network distance to nearest CGC pedestrian entrances)  0.743  20.880 0.000
ln(Distance to CBD: City Hall) 0.348 7.450 0.000 0.265 4.550 0.000
ln(Distance to nearest subway station)  0.087  30.470 0.000  0.123  34.200 0.000
ln(Distance to nearest urban arterial)  0.013  6.770 0.000  0.001  0.260 0.793

Land use and regulation

Office (0/1) 0.059 7.190 0.000 0.041 3.410 0.001
Commercial raw lands (0/1)  0.090  10.660 0.000  0.011  1.010 0.314
Mixed-use (0/1)  0.374  90.670 0.000  0.460  91.840 0.000
Mixed-use raw lands (0/1)  0.466  24.410 0.000  0.543  26.950 0.000
ln(Building coverage ratio) 0.079 4.680 0.000
ln(Floor area ratio) 0.217 58.540 0.000

Neighborhood economic and demographic attributes

ln(Population density)  0.102  3.640 0.000 0.110 4.930 0.000
ln(Employment density)  0.792  17.580 0.000  0.138  0.880 0.377
ln(Proportion of residents with college degree) 0.234 1.920 0.055 0.565 1.860 0.063
ln(Proportion of residents 40–60 years of age)  0.025  1.610 0.108 0.485 4.620 0.000
ln(Proportion of residents more than 60 years of age) 0.108 4.460 0.000 0.268 4.190 0.000

Other neighborhood attributes

ln(Park density ratio) 0.110 15.390 0.000 0.355 15.160 0.000
ln(Developed land ratio)  1.132  4.020 0.000  3.396  4.010 0.000
ln(Road area ratio) 2.946 11.870 0.000 2.196 2.520 0.012
ln(Retail area ratio) 0.065 9.890 0.000  0.021  4.630 0.000
ln(Proportion of residential permits to total permits)  0.108  19.020 0.000  0.040  7.230 0.000
ln(Proportion of commercial permits to total permits) 0.031 6.230 0.000 0.070 13.500 0.000

Constant 33.662 39.810 0.000 25.527 9.310 0.000

Random effects
ICC 0.722 0.889
Summary statistics
Number of parcel observations (Level 1) 37,515 23,969
Number of neighborhood groups (Level 2) 71 70

Fig. 13. Marginal effects of BRT bus stops on non-residential land values by distance intervals.

rose in the range of a 1.5–2.8% premium up to around 270 m stop. The absence of any nuisance effect within 30 m of a BRT stop
away. Following the BRT enhancements, premiums shot up could very well reflect the high-amenity designs of Seoul’s
significantly, eclipsing 10% up to 90 m from the nearest bus median-lane bus stops (see Fig. 15) and perhaps even a different
114 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116

Fig. 14. Marginal effects of BRT bus stops on residential land values by distance intervals.

Table 7
Multilevel hedonic model for predicting residential land value per square meter. Note: 1 Korean Won¼ 0.0011 US Dollar in 2007.

2001–2004 2005–2007

Variables Coefficient t p Coefficient t p

Fixed effects
Distance to bus stops
dummy (1, if distance r 30 m, otherwise 0)  0.025  3.710 0.000 0.103 11.050 0.000
dummy (1, if 30 o distance r 60 m, otherwise 0) 0.015 4.150 0.000 0.115 22.510 0.000
dummy (1, if 60 o distance r 90 m, otherwise 0) 0.022 8.200 0.000 0.105 26.230 0.000
dummy (1, if 90 o distance r 120 m, otherwise 0) 0.022 9.290 0.000 0.089 25.080 0.000
dummy (1, if 120 o distance r 150 m, otherwise 0) 0.026 12.020 0.000 0.082 24.980 0.000
dummy (1, if 150 o distance r 180 m, otherwise 0) 0.025 12.010 0.000 0.070 22.280 0.000
dummy (1, if 180 o distance r 210 m, otherwise 0) 0.028 13.620 0.000 0.063 20.250 0.000
dummy (1, if 210 o distance r 240 m, otherwise 0) 0.021 10.230 0.000 0.054 17.070 0.000
dummy (1, if 240 o distance r 270 m, otherwise 0) 0.016 7.520 0.000 0.053 15.920 0.000
dummy (1, if 270 o distance r 300 m, otherwise 0) 0.004 1.560 0.118 0.053 14.630 0.000

Location factors
ln(Network distance to nearest bus stop)  0.052  42.050 0.000  0.030  16.370 0.000
ln(Network distance to nearest CGC freeway ramps)  0.027  1.990 0.047
ln(Network distance to nearest CGC pedestrian entrances)  0.154  8.320 0.000
ln(Distance to CBD: City Hall)  0.004  0.250 0.800 0.081 3.380 0.001
ln(Distance to nearest subway station)  0.025  20.720 0.000  0.046  25.560 0.000
ln(Distance to nearest urban arterial)  0.049  91.310 0.000  0.044  53.010 0.000
ln(Distance to Han River) 0.007 0.900 0.366  0.379  27.000 0.000

Land use and regulation

Row Housing (0/1) 0.084 19.410 0.000 0.098 17.230 0.000
Multi-family housing (0/1) 0.041 25.850 0.000 0.051 24.520 0.000
Condominium (0/1) 0.382 80.200 0.000 0.251 58.380 0.000
Residential raw lands (0/1)  0.029  11.010 0.000  0.082  22.910 0.000
ln(Building coverage ratio) 0.167 41.250 0.000
ln(Floor area ratio) 0.105 53.970 0.000

Neighborhood economic and demographic attributes

ln(Population density) 0.162 16.110 0.000 0.095 8.040 0.000
ln(Employment density) 0.343 16.830 0.000  1.159  12.840 0.000
ln(Proportion of residents with college degree)  0.010  0.020 0.980 1.890 2.680 0.007
ln(Proportion of residents 40–60 years of age)  0.090  12.980 0.000 0.596 11.210 0.000
ln(Proportion of residents more than 60 years of age) 0.176 19.510 0.000 0.598 19.930 0.000

Other neighborhood attributes

ln(Park density ratio) 0.037 15.790 0.000 0.261 17.310 0.000
ln(Developed land ratio) 4.547 13.930 0.000  9.928  22.610 0.000
ln(Road area ratio) 5.139 43.920 0.000 12.477 16.820 0.000
ln(Retail area ratio) 0.085 38.670 0.000  0.062  33.320 0.000
ln(Proportion of residential permits to total permits) 0.027 9.000 0.000  0.083  29.590 0.000
ln(Proportion of commercial permits to total permits) 0.034 19.310 0.000 0.056 25.470 0.000
ln(CPI-adjusted local tax per households)  0.293  54.240 0.000  0.434  10.680 0.000

Constant 30.627 31.800 0.000 58.651 21.030 0.000

Random effects
ICC 0.992 0.995
Summary statistics
Number of parcel observations (Level 1) 85,124 41,302
Number of neighborhood groups (Level 2) 65 65
 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116
Fig. 15. Seoul’s high-amenity BRT bus-stop infrastructure.
Source: Seoul Metropolitan Government.
clientele who patronizes BRT than previous regular bus services.
Beyond 60 m, Fig. 14 shows that the premium effects began to
taper.
The control variables in Table 7 generally align with expectations.
Proximity to other infrastructure, including the CGC greenway,
subways, and arterial roads were associated with higher residential
land prices. Also, higher density residential parcels (e.g., row
housing, multi-family housing, condominiums) were valued more
than single-family residences (the suppressed dummy-variable
category). Factors like high neighborhood densities, high park
densities, and dense road networks also tended to increase
residential land values. The signs on some variables in Table 7,
however, are not easily interpretable and again could reflect unique,
localized attributes of Seoul’s real estate market.
7. Conclusion and policy implication
Our core research hypotheses were largely borne out by
empirical results. Seoul’s substantial upgrading of BRT services –
in the form of adding over 70 kms of dedicated median-lane bus
services in 2004 – nearly doubled bus operating speeds. In a
crowded, congested, and land-constrained city like Seoul, in-
creased accessibility prompted property owners and developers
to intensify land uses along BRT corridors, mainly in the form of
converting single-family residences to multi-family units, apart-
ments, and mixed-use projects. Moreover, land markets capita-
lized these accessibility gains, particularly among parcels used for
condominiums and higher density residential uses. Land price
premiums in the 5–10% range were estimated for residences
within 300 m of BRT stops. For retail shops and other non-
residential uses, impacts were more varied, ranging from 3% to
26% premiums over a smaller impact zone of 150 m from the
nearest BRT stop.
Hedonic price modeling is not without limitations in part
because most applications are cross-sectional and thus fail to
account for dynamic nature of bidding for sites among real-estate
suppliers and consumers (Martinez and Henriquez, 2006). The use
of data from multiple time points in our study partly responds to
such criticisms. Moreover, hedonic price models are commonly
used in tax assessments, litigation, and academic research
because over time they have demonstrated themselves to be a
statistically robust form of the more traditional sales comparison
approach of assessing land values. It might also be the case that
115
land-value premiums measured in our study were redistributive,
matched by a lowering of land-value appreciation in non-BRT-
served corridors. While we cannot rule out such possibilities, we
do not believe this was the case. This is largely because average
bus speeds and passenger throughput increased, and traffic delays
decreased, following the introduction of BRT services, despite the
continued increases in Seoul’s population and employment over
the 2001–2007 study period. These are real economic benefits
that, according to theory, should get capitalized into land values
for directly benefiting properties. And even if land-value pre-
miums are redistributive, this does not diminish the use of policy
tools that might recapture such value, such as benefit assess-
ments, to finance infrastructure improvements. Additionally,
financial redistributions can sprawn spatial redistributions –
e.g., from market pressures to build at higher densities along
BRT-served corridors – that in turn produce real benefits through
agglomeration economies (see Cervero, 1997; Boarnet and
Haughwot, 2000).
Our findings of BRT’s appreciable land-use and land-value
impacts are consistent with past work on the impacts of rail-
transit improvements. It is not transit ‘‘hardware’’ – i.e., steel-
wheel trains or rubber-tire buses – that unleash land-use changes
but rather the quality of service and more specifically, the
comparative travel-time savings of taking transit vis-a -vis the
private car. In Seoul, faster, more punctual bus services and the
ease of transferring to subway portals triggered a market demand
for higher density residential uses. Land-use intensification, along
with the access improvements conferred by BRT, also translated
into higher real-estate prices, especially for residential uses.
These research findings inform several possible policy re-
sponses. One, the desire to intensify land uses requires local
planners to get ahead of the curve by changing zoning and
regulatory restrictions governing densities and designs in advance
of BRT enhancements. This, of course, assumes higher density
development in BRT-served corridors is sought by planning
agencies. Fortunately, this is most often the case since, after all,
expensive transit investments require high ridership, which a
body of research has long shown requires high densities
(Pushkarev and Zupan, 1977; Cervero, 1998). As the saying goes,
‘‘mass transit needs mass’’. Zoning overlays, increases in permis-
sible floor-area-ratio, and density bonuses are not the only
programmatic changes that are needed in response to market
pressures to intensify uses. Other supportive infrastructure
including water and sewerage trunk-line capacities have to be
116 R. Cervero, C. Kang / Transport Policy 18 (2011) 102–116
upgraded and expanded to serve more households and busi-
nesses. Linking infrastructure like BRT to local zoning and land-
use planning seems fairly straightforward; however, it should be
remembered that many cities in the developing world aiming to
economize on transit investments by building BRT (e.g., Jakarta,
Ankara, Cali, Abidjan) do not always have the institutional
capacities and resources to carry out strategic land-use planning.
As noted earlier, the presence of measurable land-value
premiums conferred by BRT create revenue-generating opportu-
nities as well, notably transit value capture. Since BRT is a public
investment that yields benefits to private property owners, value
capture aims to return a share of the value-added to public coffers
to help finance the capital investment and subsequent opera-
tions. BRT-induced land appreciation can be partly recaptured
through benefit assessment district financing and public–private
joint development initiatives. Presumably, all parties in joint
development deals perceive benefits from intensifying develop-
ment near transit – i.e., increased ridership for public operators
and increased land prices and rents for private landholders –
which eases the process of hammering out revenue-sharing
agreements. Another implication of rising land prices, of course,
is displacement of lower-income households and other potential
mal-distributive effects. To redress such inequities, one possible
use of revenues recaptured from benefitting property owners is to
underwrite the costs of providing affordable housing and shops to
displaced residents and merchants.
It should be kept in mind that Seoul’s BRT improvements did
not occur in isolation. Rather they were part of a larger campaign
to reclaim land given over to freeways and to enhance urban
living as an alternative to exurban new-town development. No
project epitomized this shift in urban policy more than the
freeway-to-greenway conversion, Cheong Gye Cheon (CGC).
Indeed, former mayor Myung-Bak Lee opted to upgrade the city’s
bus services and invest in dedicated-lane BRT partly to ensure
high-quality transit was in place to absorb capacity lost and trips
deflected by the CGC freeway demolition. In Seoul, BRT improve-
ments were part of a larger policy agenda that required a systems
approach. While our research focused on median-lane bus
services, some of the estimated benefits were no doubt tied to
other initiatives introduced at roughly the same time to improve
mobility and quality-of-living in central Seoul. As is often the case,
this one infrastructure component – BRT – was likely a necessary,
though by itself, insufficient, factor in intensifying residential
activities and increasing land values. Seoul’s experiences under-
score the importance of applying a systems approach to transit
investments, tied to a larger public purpose, which in the case of
Seoul included re-generation and revitalization of the urban core.
Acknowledgements
We are grateful to OPENmate, a GIS company based in Korea,
for technical support and to Dr. Gyeng-Chul Kim from Veolia
Transport Korea for assistance with this research.
References
Bae, C., Jun, M., Park, H., 2003. The impact of Seoul’s subway line 5 on residential
property values. Transport Policy 10 (2), 85–94.
Banister, D., 2005. Unsustainable Transport: City Transport in the New Century.
Routledge, London.
Banister, D., Lichfield, N., 1995. The key issues in transport and urban development.
In: Banister, D. (Ed.), Transport and Urban Development. E&FN Spon, London.
Bartik, T., 1988. Measuring the benefits of amenity improvements on hedonic
models. Land Economics 64 (2), 172–183.
Bhatta, S., Drennan, M., 2003. The economic benefits of public investment in
transportation: a review of recent literature. Journal of Planning Education and
Research 22, 288–296.
Boarnet, M., 1998. Spillovers and the locational effects of public infrastructure.
Journal of Regional Science 38 (3), 381–400.
Boarnet, M., Chalermpong, S., 2001. New highways, house prices, and urban
development: a case study of toll roads in Orange County, CA. Housing Policy
Debate 12 (3), 575–605.
Boarnet, M., Haughwot, A., 2000. Do Highways Matter? Evidence and Policy
Implications of Highway’s Influence on Metropolitan Development. Brookings
Institution Center on Urban and Metropolitan Policy, Washington, DC.
Cambridge Systematics, Cervero, R. and Aschuer, D., 1998. Economic Impact
Analysis of Transit Investments: Guidebook for Practitioners. National
Academy Press, Washington, DC. Transit Cooperative Research Program,
Report 35, National Research Council, 1998; with Cambridge Systematics
and D. Aschuer.
Cervero, R., 1997. Transit-induced accessibility and agglomeration benefits: a land
market evaluation. Working Paper 691, Institute of Urban and Regional
Development, Berkeley.
Cervero, R., 1998. The Transit Metropolis: A Global Inquiry. Island Press, Washington.
Cervero, R., 2004. The property value case for transitIn: Dunphy, R. (Ed.),
Developing Around Transit: Strategies and Solutions that Work. Urban Land
Institute, Washington, DC.
Cervero, R., Landis, J., 1997. Twenty years of the bay area rapid transit system: land
use and development impacts. Transportation Research Part A 31 (4), 309–333.
Dowall, D., Monkkonen, P., 2007. Consequences of the Plano Piloto: the urban
development and land markets of Brasilia. Urban Studies 44 (10),
1871–1887.
Du, Hongbo, Mulley, C., 2006. Relationship between transport accessibility and
land value. Transportation Research Record 1977, 197–205.
Estupinan, N., Rodriguez, D., 2008. The relationship between urban form and station
boardings for Bogotá’s BRT. Transportation Research Part A 42, 296–306.
Ewing, R., 2009. Transportation infrastructure and urban growth and development
patterns. In: Boarnet, M. (Ed.), Transportation Infrastructure: The Challenges of
Rebuilding America. American Planning Association, Planning Advisory
Service, Chicago Report Number 557, pp. 27–39.
Gatzlaff, D., Smith, M., 1993. The impact of the Miami metrorail on the value of
residences near station locations. Land Economics 69 (1), 54–66.
Giuliano, G., 2004. Land use impacts of transportation investments: highway and
transit. In: Hanson, S. (Ed.), The Geography of Urban Transportation. The
Guilford Press, New York, pp. 237–273.
Hall, P., 1995. A European perspective on the spatial links between land use,
development and transport. In: Banister, D. (Ed.), Transport and Urban
Development. E&FN Spon, London.
Huang, H., 1996. The land-use impacts of urban rail transit systems. Journal of
Planning Literature 11 (1), 17–30.
Levinson, H., Zimmerman, S., Clinger, J., Rutherford, S., 2002. Bus rapid transit: an
overview. Journal of Public Transportation 5 (2), 1–30.
Martinez, F., Henriquez, R., 2006. A random bidding and supply land use
equilibrium model. Transportation Research B 41, 632–651.
McMillen, D., McDonald, J., 2004. Reaction of house prices to a new rapid transit line:
Chicago’s midway line 1983–1999. Real Estate Economics 32 (3), 463–486.
Perez, P., Martinez, F., Ortuzar, J., 2003. Microeconomic formulation and estimation
of a residential location choice model: implications for the value of time.
Journal of Regional Science 43 (4), 771–789.
Polzin, S., Baltes, M., 2002. Bus rapid transit: a viable alternative? Journal of Public
Transportation 5 (2) 47–69.
Pushkarev, B., Zupan, J., 1977. Public Transportation and Land Use. Indiana
University Press, Bloomington.
Rabe-Hesketh, S., Skrondal, A., 2008. Multilevel and Longitudinal Modeling Using
STATA. STATA Press, College Station, TX, USA.
Raudenbush, S., Bryk, A., 2002. Hierarchical Linear Models: Application and Data
Analysis Methods. Sage, Newbury Park, California.
Royal Institution of Chartered Surveyors, 2002. Land Value and Public Transport.
Royal Institution of Chartered Surveyors, London.
Rodriquez, D., Targa, F., 2004. Value of accessibility to Bogotá’s bus rapid transit
system. Transport Reviews 24 (5), 587–610.
Rodriquez, D., Mojica, C., 2008. The land value benefits of bus rapid transit: the
case of Bogotá’s transmilenio. Landlines 24, 2–6.
Rosen, S., 1974. Hedonic prices and implicit markets: product differential in pure
competition. Journal of Political Economics 82, 34–55.
Ryan, S., 1999. Property values and transportation facilities: finding the transporta-
tion—land use connection. Journal of Planning Literature 13 (4), 412–427.
Seoul Development Institute, 2005a. Toward Better Public Transport, Seoul
Development Institute, Seoul Metropolitan Government.
Seoul Development Institute, 2005b. Monitoring Service Level of Seoul Bus System.
Seoul, Korea: Seoul Development Institute Seoul, Seoul Metropolitan Govern-
ment.
Seoul Metropolitan Government, 2009. Achievement and Present State of Public
Transit Reform in Seoul, 2009. Seoul Metropolitan Government, Seoul.
Stigler, S., 2008. Fisher and the 5% level. Chance 21 (4), 12.
Voith, R., 1993. Changing capitalization of CBD-oriented transportation system-
s—evidence from Philadelphia, 1970–1988. Journal of Urban Economics 33 (3),
361–376.
Vuchic, V., 2002. Bus semirapid transit model development and evaluation. Journal
of Public Transportation 5 (2), 71–96.
Zhou, B., Kockelman, K.M., 2008. Neighborhood impacts on land use change: a
multinomial logit model of spatial relationships. The Annals of Regional
Science 42, 321–340.